skill-router · dotslashsimran — Claude Code skill for choosing and chaining skills
The skill-router Claude Code skill selects a matching skill or builds a sequence for a task. It kicks in when a request is vague or could call for several skills.
MIT ★ 541 updated 2026-11-09
npx skills add dotslashsimran/simcraft --skill skill-router -a claude-codeWhat skill-router does
skill-router compares the task with installed skill descriptions, checking whether each candidate fits the trigger, available input, and expected output. It discovers skills by running catalog.py instead of relying on memory. The script requires python3.
For multi-stage work, the router passes each stage’s output to the next skill and caps the chain at three. It presents a short plan before work begins. If an unclear goal, input, or deliverable would change the choice, it asks one question; if no skill fits, it handles the task directly.
What it helps with
- choose a skill for a vague task
- match installed skills to an input and expected deliverable
- connect distinct stages through explicit handoffs
- avoid chaining skills that do the same job
Example prompts for the agent
- Choose the right skill for this task and outline the steps.
- Find skills for each stage, then run them in order.
- Check whether any installed skill fits, and do the task directly if none does.
How to install skill-router
1Skills CLI (Vercel)
npx skills add dotslashsimran/simcraft --skill skill-router -a claude-code2Manually: put the files in place
~/.claude/skills/skill-router/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/skill-router/
1Skills CLI (Vercel)
npx skills add dotslashsimran/simcraft --skill skill-router -a codex2Manually: put the files in place
.agents/skills/skill-router/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add dotslashsimran/simcraft --skill skill-router -a cursor2Manually: put the files in place
.cursor/skills/skill-router/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install dotslashsimran/simcraft skills/planning/skill-router2Manually: put the files in place
.github/skills/skill-router/SKILL.md1Skills CLI (Vercel)
npx skills add dotslashsimran/simcraft --skill skill-router -a gemini-cli2Platform CLI
gemini skills install https://github.com/dotslashsimran/simcraft --path skills/planning/skill-routerManually: put the files in place
.devin/skills/skill-router/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about skill-router
How do I install skill-router in Claude Code?
Run npx skills add dotslashsimran/simcraft --skill skill-router -a claude-code. The skill file is at ~/.claude/skills/skill-router/SKILL.md.
What does skill-router do?
It matches a task against installed skill descriptions and chooses one skill or a chain of up to three. It shows a plan first and passes each stage’s output to the next.
What happens if no installed skill fits the task?
The router handles the task directly instead of forcing a match. It scans the skill catalog by running catalog.py with python3.
Source text
SKILL.md · MIT
---
name: skill-router
description: Decide which installed skill, or which chain of two or three skills, fits a task before starting it, and run them in the right order with each one's output feeding the next. Works with any set of skills, because it reads every installed skill's description at run time. Use when a task is big or vague, could fit several skills, or needs more than one ("which skill should I use", "what's the best way to do this", "plan the steps", "use whatever skills make sense"), or when it's unclear whether any skill applies.
---
# Skill router
Pick the right tool before working. One skill done well beats three loosely chained, so prefer one skill, use two when the task truly has two stages, and use three only for a full pipeline.
## 1. See what's installed
```
python3 ~/simcraft/skills/planning/skill-router/scripts/catalog.py # every skill, with its description
python3 ~/simcraft/skills/planning/skill-router/scripts/catalog.py --grep video
```
Always run it rather than relying on memory. Skills get added and removed, and the description is the contract.
## 2. Pin down the task
Write one line each for:
- **Goal:** what the user wants to be true at the end.
- **Input:** what exists now (a file, a URL, a repo, a vague idea).
- **Deliverable:** the exact artifact (a video, a report, a fixed bug, a doc).
If any of these is unclear and the answer would change the skill choice, ask one question. Otherwise decide.
## 3. Match
For each candidate skill, check three things against its description:
1. **Trigger fit.** Does the task match what it says it's for, not just share a keyword?
2. **Input fit.** Does it accept what the user has?
3. **Output fit.** Does it produce the deliverable, or something the next skill needs?
Keep only skills that pass all three for their part of the work. When two skills overlap, choose the more specific one (a skill made for the exact format beats a general one). If nothing fits, say so and do the task directly. A forced match is worse than none.
## 4. Decide: one, two or three
- **One skill** when it goes from the input to the deliverable on its own.
- **A chain** when the work has distinct stages and each stage has its own best skill. Typical shapes:
- **understand → make:** analyse something first, then build from what was learned.
- **make → check:** build, then audit or review it with a separate skill.
- **raw → clean → finish:** e.g. trim footage, clean the audio, then add graphics.
- Write the chain as a handoff: what each skill produces, and which file or fact the next one takes as input. If a step has no clear handoff, the chain is wrong.
- Never run two skills that do the same job, and never add a skill just because it is related.
## 5. Show the plan, then run
Show the user a short plan before starting anything heavy:
```
Task: <one line>
Plan: <skill A> (<what it gives>) → <skill B> (<what it does with it>)
Why not <closest alternative>: <one line>
```
Then load each skill with the Skill tool in order, follow its instructions fully, and check each handoff before moving on. If a stage's output is weak, fix it there rather than letting the next skill build on it.
## Examples of good chains
These use skills from this repo. The method is the same for any skill set.
| Task | Plan |
|---|---|
| "Remake this viral clip in my style" | reverse-engineer-video (shot list, camera moves, beat map) → perspective-illusion or code-loop (build it) |
| "Turn this paper into a video" | paper-to-runnable (working code + internals animation) → vox-explainer (explainer over a voiceover) |
| "Audit this SDK and show it" | dx-audit (friction log, teardown) → scripted-screencast (recorded journey) |
| "Clean up my talking-head take" | rough-cut → clean-audio |
| "Is this new endpoint safe and well designed?" | api-design-review → security-threat-model |
| "It got slower last week" | benchmark-honestly (a trustworthy oracle) → bisect-anything (find the commit) |
| "I just joined this repo, fix the flaky CI" | codebase-map → flaky-test-killer |
| "Make a branded explainer for a sponsor" | brand-kit → vox-explainer |
## Mistakes to avoid
- Picking by keyword ("video" in the request) instead of by deliverable.
- Chaining when one skill already covers both stages.
- Skipping a stage's own checks to get to the next skill faster.
- Forgetting a skill's safety rules when it runs inside a chain. Each skill's rules still apply in full.
© dotslashsimran · MIT · dotslashsimran/simcraft
Similar skills
All skills for Agent setup: memory, context, orchestration →skill-router by other authors comes first.